J. Phys. D Appl. Phys. 27 (1994) 1757-1762.

Printed in the UK

Cathodoluminescenceand depth
profiles of tin in float glass
B Yangt,

P D Townsend and S A Holgate

MAPS, University of Sussex, Brighton BN1 9QH, UK

Received 9 March 1994

Abstract. Cathodoluminescence (CL) spectra have been recorded from float glass
samples and the data used to determine the depth profile of tin introduced from
the float process. The luminescence data were interpreted in terms of intrinsic
emission bands at 380 and 410 nm and well resolved extrinsic emission bands
resulting from the tin impurities at 575 and 630 nm. Signals were recorded at
temperatures from 40 to 300 K. The presence of the tin produces an optical
waveguide in the glass surface as a result of the float production process. It
was therefore possible to depth profile the tin by measurement of the waveguide
modes. The CL and waveguide methods are in excellent agreement. Overall, the
depth of tin diffusion was up to 8-9 Nm in the samples studied. The lifetimes
and intensity of the signals are temperature dependent, and these parameters
are reported for the 380 nm and 575 nm emission lines. Thermal treatments
modify the CL spectra and such changes are considered the result of valence or
aggregation states of the tin impurities.
1. Introduction

Whilst float glass is a familiar material for building

applications, it should be recognized that there are
some interesting properties which have not been
exploited or studied in detail for other applications.
Foremost among these is that the float process inevitably
introduces tin, by thermal diffusion, into one face of
the glass 113. The presence of the tin raises the
refractive index and consequently forms a very low-loss
planar waveguide [2,31. Applications for such large area
planar waveguides have not yet been publicly proposed.
At a more fundamental level it is worth enquiring if the
concentration profile of the tin and the refractive index
changes are the same. Further, one should consider if the
tin is in the optimum valence state for such changes in
refractive index, and whether one can control the state
of incorporation of tin within the glass. The present
paper addresses these questions using the two totally
independent techniques of waveguide mode profiling and
cathodoluminescence spectral measurements.

2. Experimental details
Measurements were made on float glass samples from
both Societa Italiana Vetro (SIV) and from Pilkington
plc. Since the luminescence signals were very similar,
only the data from SIV glass are reported here. The
samples were cut into 10 x 10 mm2 pieces and cleaned

before examination. The thickness of these samples

was 3 nun. For the cathodoluminescence, an electron
beam of energy 10 keV was focused onto a spot of
approximately 2 mm in diameter. Beam currents of
0.03 PA were used to excite the samples. Since
changes in beam energy and current density might
modify the emission spectra and intensity, for the present
paper all the spectra shown here were recorded with
the same beam conditions. The electron beam was
chopped at a frequency of 90 Hz, except for the lifetime
measurements, and the signal output was measured with
a lock-in amplifier.
Spectra were measured in the temperature range 40
to 300 K using an f/4 scanning monochromator and a
cooled red-sensitive photomultiplier tube. All data were
recorded on a computer and corrected for the spectral
response of the whole detection system, as described
earlier [4]. A spectral resolution of 10 nm was used.
Depth profiles using the optical waveguide properties
of the tin side of the glass were made by measuring the
waveguide mode spacing with both a 632.8 nm HeNe
laser and a 488 nm argon beam by the standard dark
mode prism coupling method [SI. The depth profile was
them determined using a reflectivity analysis described
by Lama and Chandler [6].
3. Results

3.1. Optical waveguide data and depth analysis

t Permanent address: Physics Department, Beijing Normal

University, Beijing 100875, China

The float process produces flat glass by allowing

the liquid glass to spread across a sheet of molten

0 1994 IOP Publishing Ltd

1757

0022-3727/94/081757+6$19.50

B Yang et a/
1.5301

1.5281
X

1.5281

4
0

.k

1.5211

&

2 1.5281
1.5251
0. 0

2.00

4.00

d o

Depth (Microns)

Figure 1. A comparison of the refractiveindex

and cathodoluminescencechanges produced by
tin in a float glass sample. The full cuwe is the
index profile determined from optical mode analysis.
Cathodoluminescence data are shown after several
polishing steps.
tin. During this hgh-temperature phase of the glass
production tin diffuses into the lower face of the glass to
a depth of several micrometres. There is also some tin
contamination of the upper face of the glass from the tin
vapour, although this is in much smaller concentrations.
The upper face tin content is sensitive to the furnace
conditions and so vanes not only with the supplier but
also with the glass thickness and flow rate across the
tin bath. The two faces are readil) distinguishable
since the tin-rich side fluoresces when exposed to UV
illumination. Less obviously, one may detect the change
in refractive index and surface reflectivity resulting from
the tin impurities. Since the tin raises the refractive index
of the glass, the layer is potentially an optical waveguide
12.31. Indeed such glass layers providc extremely low
cost, very large planar waveguides of excellent quality
and low optical loss [2]. To a first approximarion one
may assume that the refractive index enhancement within
the waveguide is proportional to the concentration of the
tin impuriry ions. In order to obtain the diffusion profile
of tin in the surface of the glass, the following method
has hecn used. The conventional waveguide profiling
method is to couple a laser beam into the waveguide via
a prism. Measurement of the critical angles at whch
the mode indices within the guide match the coupling
reveal the mode spacings within a multimode guide.
Such data were obtained with both a red HeNe laser at
632.8 nm and a blue argon line at 488 nm. The profile
shapes were determined by waveguide mode analysis,
although there are only a limited number of measurable
modes, with a maximum of four modes, even for the
shorter wavelength. Nevertheless, one can estimate the
refractive index profilc from thc mode spacing, and
compare the profiles from the two sets of wavelength
data. In practice these were in excellent agreement.
They lead to the profile of the form shown by the
continuous line in figure 1. The profile shows that there
is initially a saturation value, to a depth of 2.95 pm.
This is followed by a tal in the refractive index profile
1758

which is of a Gaussian shape, as expected for a diffusion

process. Overall the tin penetration extends to beyond
8 pm for the example chosen.
Since the mode spacings are dependent on the index
profile, and thus the depth of the waveguide, the mode
values may be used to monitor the effects of sectioning
of the guide as surface layers are successively removed
by polishing. Using fine diamond power of 1 p m
diameter, the Sn surface of the sample was polished, step
by step, so as to remove the tin diffused layer. As the
tin diffised glass is a planar waveguide the movement of
the mode indices are a function of guide thickness, and
so determine the extent of the residual tin layer. The
multimode measurements was adequate so long as the
layer was thicker than 5 pin. After some practice, a
linear relationship between the thickness polished from
the layer and the polishing time was established. The
polishing rate was constant with time, and so it was
possible to predict the thickness of the residual doped
layer even when the layer was thinner than 5 pm.
(This information on polishing rate was required for the
parallel experiment using cathodoluminescence.) The
refractive index profiles determined after each polishing
step were in excellent agreement, and the profile curves
for the various residual thicknesses defined the same
form for the index changes for the inner part of the
diffused glass.

3.2. Cathodoluminescence data

The cathodoluminescence emission spectrum and intensity are not easily interpreted in the case of a nonuniform dopant situation, since the rate of energy deposition from the electron beam, is a function of depth and,
at least in principle, the efficiency or light production and
the spectrum can be sensitive fo the rate of excitation.
There are therefore several problems to address in using
CL to monitor the presence of impurities in a quantitative fashion. The first is to ascertain if the spectrum has
a signal which is uniquely linked to the impurity, the
second is to assume that the signal is linearly dependent
on the dopant concentration in the glass, and insensitive
to the rate of electron beam energy deposition. For float
glass the Sn impurity is clearly dominant on the side
which was in contact with the tin. Nevertheless, some
tin reached the upper face by vapour transport. However, tin is unlikely to reach the interior. Hence some
of the SW glass was cut to provide thinner pieces of
10 x 10 x 1.5 mm3. The objective of such a cutting
operation was to provide three distinct types of surface.
One, the face labelled Sn corresponds to the side of the
glass in contact with the molten tin during the float production. The opposite (i.e. the non-tin) side, labelled
N, has become lightly doped with tin vapour. The inner faces, labelled I, exposed by cutting the sample, are
assumed to be essentially free of tin.
The CL spectra of these three surfaces are distinctly
different and are shown in figure 2. The following
features can be deducted from this figure.

Cathodoluminescenceand depth profiles ot tin in float glass

( I ) Two peaks at 380 and 410 nm always exist in

the luminescence from interior sections of the glass, and
so are almost certainly the result of intrinsic emission
processes. These intrinsic emission bands have a long
wavelength tail extending over all the visible range. The
intensity of the tail decreases with longer wavelengths
and contains several component bands, as discussed in
a later section.
(2) The strong peak at 575 nm and the broad
shoulder peaked near 630 nm undoubtedly result from
tin emission bands, because these bands are extremely
strong in the spectrum of the Sn surface, and do
not appear in that of the I surface. Two types of
additional measurements confirm this. The first is that
the intensity of the signals from the so called non-tin side
is approximately linear with glass thickness and this is
closely related to the time spent in the float section of
the furnace. Data from glass coated with a deposit of
SnOz produced broadly the same strong features. The
575 nm peak was still dominant but some of the weaker
features differed slightly from that of the float glass.
In particular the 630 MI broad band in the float was
split into two components near 610 and 645 nm and the
shorter wavelength peaks were better resolved. Overall
these results conlimn the role of the tin but suggest that
a number of emission bands may be hidden with the
envelope described here.
(3) Weaker components on the short wavelength side
of the Sn band (e.g. near 460 and 515 nm) are noted but
it is difficult to decide whether they relate directly to the
tin, intrinsic sites, or other luminescence sites formed
for example as intrinsickin complexes. In the data for
the non-tin side versus glass thickness the intensity of
these bands relative to the 575 nm Sn signal varied in
the same way as did that for the intrinsic 380 nm signal.
This suggests that these smaller features are not primarily
from transitions at tin sites.
(4) As mentioned in (2),in the CL spectrum of the
upper (N) surface, the tin bands were also detected,
although their intensities are much lower than those from
the face in contact with the tin. They were also variable
in intensity and the ratio of the component bands was
not constant in all samples. It can be estimated that the
tin concentration in the N surface is about 16% of that
in the Sn surface for the 3 mm thick samples (assuming
a linear intensity relationship between the CL signal and
the tin concentration). The intensity of the tin signal
from the upper face is approximately proportional to the
time spent during the passage of the glass over the tin
zone.
3.3. Cathodoluminescenceand depth analysis

On the float side of the glass, with a high tin

concentration, it is possible to compare the refractive
index profile with the CL intensity. In order to avoid
secondary effects from CL depth profiling using variable
electron energy it is preferable ro use fixed electron beam
conditions, and selectively remove material from the
glass surface by polishing. One notes that the electron

36000

30000

41

\
6000

.-._
.---______
300

400

500
6 o
700
Wavelength (nm)

aoo

900

Figure 2. A comparison of the CL spectra from the tin side

(Sn),the non-tin side (N) and the interior (I) of float glass.
probe beam has an energy against depth excitation profile
which is very approximately that of a Gaussian line
shape, and the maximum electron range of IO keV
electrons in glass is near half a micrometre. However,
since the glass surface is angled relative to the ion beam
in order to record the emission spectra, the effective
CL analysis is only from a depth of some 0.3 p m .
Deconvolution of the electron energy deposition profile
and the depth dependence of the signal on a finer
depth scale was thus not attempted. For the fixed
beam condition the CL spectrum was recorded after
every polishing step, and compared with the waveguide
analysis. CL experimental conditions are difficult to
reproduce precisely and so a signal normalization was
used in which the 575 nm emission, from tin, was
compared with that near 380 nm, which is considered
to be an intrinsic signal of the glass host. The 575 nm
luminescence signal is assumed to be proportional to
the tin concentration and further, one assumes that the
presence of tin does not influence the efficiency of light
production at 380 nm. During profiling the intensity data
of the 575 nm emission relative to that at 380 nm are
superposed on the waveguide data of figure 1. The form
of the index and CL profiles are in good agreement. The
linearity of the polishing rate allows the CL analysis to be
extended into the tail of the index profile, even beyond
the point when mode values offered depth information.
The important point to note is that the two totally distinct
approaches to the tin profiling offer the same depth
penetration profiles for the tin.

3.4. Thermal treatments

During glass production the material cools whilst in the
form of a massive sheet of glass. This cooling phase
can influence the manner in which the tin is incorporated
into the glass network. Alternative possibilities include
different degrees of coalescence of the tin, for example
as colloidal precipitates, or as tin in different charge
states [7]. Colloidal metal precipitation is probably
unimportant, since there was very low loss and scattering
in the earlier waveguide loss data 121. However, the
1759

B Yang et a/
26000,

30000

._..._..SLOW coomc

FAST COOLlNG
ROOK TEMPEKATURE

20000

2
380nm
575nm
.

0
0

Wavelength (nm)

5b

Id0

150

200

250

300

Temperature (K)

Figure 3. Changes induced in the w spectra from the

non-tin side after heating to 600 "Cand either cooling
quickly or slowly.

Figure 5. CL intensily of the 575 nm and 380 nm bands of

float glass as a function of temperature.

3.5. Lifetime and temperature dependence

I
04

zoo

I
.',

zoo

,?'--

ddo

.500

I
so0

760

Wavelength (nm)

abo

d o

Figure 4. An example of CL spectra from an interior glass

surlace before and after fast cooling from 600C.
identification of altemative valence states of the tin is
a more difficult problem. Assuming that changes in
valence might be reflected by variations in intensity or
luminescence spectra, an attempt was made to modify
the spectra by thermal treatments. With this objective
in mind the glass samples were heated to various
temperatures up to 600 "C, and then either slowly cooled
(down to 300C in about 5 m h ) or rapidly removed
from the furnace and placed on a cold metal surface.
In principle this could establish and stabilize different
degrees of imperfection, or valence, for impurities or
intrinsic defects in the glass. In practice the CL data from
the tin side were insensitive to the various treatments, but
both the non-tin and interior faces gave signals which
were obviously modified. The pattern of changes is
complex, as it depends on both the maximum furnace
temperature used and the cooling rate. In order to
demonstrate the scale of the changes, examples are given
in figure 3 for a 'fast' and 'slow' cooled glass recorded
from the non-tin side. Figure 4 shows similar data for
interior faces of the glass. The maximum temperature
used for these examples was 600C.
1760

Intensities of the 380 nm and 575 nm CL emission

bands of the Sn surface of SIV glass were recorded at
various temperatures from 300 down to 40 K and the
results are shown in figure 5. One may characterize
such temperature variations by an activation energy
E , by assuming that there is a temperature insensitive
component plus a thermally activated element to the
signal. Effectively this assumes that there is a thermally
activated balance between the decay routes for radiative
and non-radiative emission. This evaluation is based
on the fact that in the higher temperature range the
additional emission intensity A I is proportional to
exp(-Elk?'),
where AI is equal to (IO - I ) , and Io
and I are the maximum intensity and the intensity at
temperature T of the same emission band respectively.
The relationships between In(AI) and 1/T of the
380 nm and 575 nm bands are linear over the temperature
range 267 to 188 K. The gradients of the linear regions
correspond to values of E of 0.046 eV and 0.064 eV for
the 380 nm band and 575 nm bands respectively.
Additionally, the phase difference qi between the
reference and signal can be measured by using the lockin amplifier. Hence, in principle, the signal lifetime
r can be obtained from the relation tan4 = &us,
where U is the chopping frequency. Unfortunately in
the present CL system the maximum chopping frequency
is only 100 kHz and this is slow compared with the
luminescence lifetimes being considered. Nevertheless,
attempts to estimate the lifetime of the 380 nm intrinsic
emission gives a value near 0.8 x IOb6 s at mom
s at 40 K. The
temperature (RT), and near 1.0 x
values for the 575 nm tin band are probably longer,
with values near 1.1 x
s at RT, and 1.2 x IOm6 s
at 40 K. As the lifetimes of all the emission bands are
similar to each other, and too short for the present CL
system, it was not possible to improve the resolution
of the overlapping bands by using different chopping
frequencies.

Cathodoluminescenceand depth profiles of tin in float glass

4. Discussion
The immediately obvious feature from figure 1 is that
the independent monitors of the effects of tin diffusion,
namely via the refractive index and the longwavelength
CL provide the same depth profile. One can thus
attribute such changes to the penetration of tin into
the glass. The data suggest that there is a saturation
region near the surface followed by a Gaussian-type
diffision tail extending down to some 8 p m depth.
Intuitively this is a reasonable consequence of the float
process. However one should note that the analyses
are on a relatively coarse depth scale and thus smaller
scale variations are smoothed over. Earlier analyses of
the depth concentration of the Sn 17-11] using SIMS,
RBS, Auger and Mossbauer techniques are not totally
consistent. They generally show an overall pattern of a
decreasing concentration with depth, but in some cases
there are indications of a peak in the concentration close
to the surface (atdepths of, say, 100 nm). Such subtleties
are not resolvable by the current techniques.
In order to analyse the luminescence data one should
note that the SIV glass contains more than 70% silica, and
some alkali and alkaline earth oxides, e.g. NazO, CaO
and MgO. For such a glass the atomic structure has a
short-range regular arrangement of atoms but a complete
lack of any long-range periodicity and regularity. The
block is the most important
silica tetrahedron
molecular building unit and the glass lattice network is
fomed from such units, but the oxygen ions may also be
linked to an alkali metal or an alkaline-earth metal ion.
The intrinsic emission from the silicate group has been
studied in many types of sample, including such diverse
examples as silica and quartz, silica fibres, feldspars
and zircon (e.g. see reviews in [4,12-141). Despite
the wide selection of host structnres there are many
similarities with a general pattern of blue/UV emission
in the wavelength range 300 to 380 nm, which is
variously ascribed to intrinsic luminescenceemitted from
relaxed self-trapped excitons decaying at the intrinsic
(SO4) site, or wavelength shifted variations of this
recombination resulting from the presence of defects
such as the intrinsic examples of E or E, centres, or from
common impurities such as H, Al or Ge. Consequently,
for the present data, the assumption is that the emission
bands near 380 nm and 410 nm in the blue end of the
spectrum of this silicate glass are similarly related to
intrinsic electron hole recombination. In some of the
samples the 380 nm signal was more pronounced after
heat treatment. Since there are considerable variations
in the assignments of the origin of the defects which
produce luminescence, even at the same wavelength,
table 1 lists some of the more common interpretations.
CL is not a passive probe of the defect sites and,
whilst changes in energy obviously sample different
depths of material, variations in current at the same
electron beam energy may alter the relative intensities
of different emission bands. In part this is because
for an insulator the secondary electron emission and
space charge development induce a complex electric

Table 1. Examples of emission bands seen in silicate

materials and tentative assignments of their origin.
Band position

(nm)

(eV)

Method

Possible model

260
290
370
380
390
400
450-480
470
570
580
640

4.77
4.28
3.78
3.25
3.17
3.1
2.5-2.8
2.64
2.18
2.15
1.9

CL

pL,cL,n

Oxygen vacancy
Oxygen vacancy

RbTL,

Ha04

RL
PL,TL
RL,CL

Seen in ultrapure quartz

Oxygen vacancy
O2 intrinsidGe impurity
Self-trapped exciton

w,n

A104

a,cL,n

CL
CL
CL.16

Ge impurity
Oxygen vacancy
Na impurity

Note: Oxygen vacancy covers alternative versions of the

structural defects such as E, E;, etc. These bands
have been reported for at least those methods shown
based on recent literature. CL, PL, RL, TL are cathodo-,
photo-, radio- and thermoluminescence respectively. IB is
ion beam stimulated.
field distribution in an insulating glass [15]. The scale
of the changes in surface potential may be comparable
to the primary electron beam energy. Surface states
can modify the alternative radiative and non-radiative
decay routes of the excited material and hence alter the
intensities of different emission bands. As evidence
for such effects, figure 6 shows the CL intensity of
380 nm and 575 nm bands as a function of the electron
beam current. Note that since both signals are recorded
together from the same sample the differences in CL
intensity are not simply experimental artefacts. One
notes that the relationship between the intensity of
the 380 nm emission and the beam current is slightly
super-linear. Numerous factors might play a role
in these similar, but differing, responses of the two
main emission bands confirm that they are of distinctly
different origin. For example, we may speculate that the
380 and 575 nm emissions are maximized at different
depths within the electron penetration range, and so are
variously influenced by surface charging and internal
electric fields. Such factors would result in a changed
sensitivity to current density or local beam heating.
Such a suggestion is prompted by the knowledge that
the surface composition of glass frequently differs from
that of the interior, particularly as a result of alkali ion
migration and surface penetration of water.
When the tin atom, with the electronic configuration
5Sz, P2,diffuses into the glass, electron transfer can
affect one or more electrons. So the tin may be
incorporated in the glass as an interstitial atom, or as
Sn linked with either one or two oxygen ions. Earlier
studies using Mossbauer spectroscopy indicate that the
ratio of SntZ to Snt4 states is a function of depth and
production conditions 171. One assumes that each type
of ion site can produce characteristic emission features
which, in principle, may be associated with the observed
tin emission bands observed in the present work. The
heat treatments modified the spectra on the non-tin side
1761

B Yang et a1

opposite glass face, doped by vapour transport is not in

equilibrium and subsequent thermal processing can alter
the state of the tin within the glass network.

12000

Acknowledgments

We are grateful to the Royal Society for financial

support. The work is also part of the EEC BRITE
E
W programme, contract BREU-0357. We further
wish to thank Societa Italiana Vetro for encouragement
and helpful discussions.

4000

2oooj
0

,,io

o.io

040

0.60

0.40

0.50

0.60

0.70

Beam Current (PA)

Figure 6. CL intensity of the 575 nm and 380 nm emission

bands of float glass against electron beam current at room

temperature.
of the glass but on the float side the changes were
minimal, implying that at the high tin dopant levels the
glass had reached a saturation equilibrium. This view is
confirmed by the refractive index data where the index
is constant for nearly 3 pm into the glass.

5. Conclusion

In conclusion one notes that the introduction of tin into

float glass directly influences the refractive index and
the CL intensity. Emission bands near 575 and 630 nm
are associated with the tin. The float process leads to
a saturation of the signals near the glass surface which
was in contact with the molten tin. One assumes that
tin diffuses into the glass in the molten phase and the
fact that both the index and CL signals show a surface
saturation, followed by a Gaussian tail deeper in the
glass, suggests that there is a linear correlation between
the tin concentration and the observed changes. The

1762

References
[11 Pikington L 1969 Proc. Roy. Soc. A 314 1
121
. . Kakarantzas G. Glavas E and Townsend P D 1989
Electron. Lek. 25 102
131
_ .Gevorevan S Sh. Zabunvan M H and Vrtanessvan G S
198yOpr. Q&t.
El&tron. 21 307
141 Luff B J and Townsend P D 1990 J. Phys.: Condens.
Marter 2 8089
1-51 7Ien P K, ulrich R and Martin R 1 1970 J. Opt. Soc. Am.
60 1325
[61 Lama F L and Chandler P J 1986 Opt. Acta. 33 127
[71 Principi G. Maddalena A, Gupta A, Geotti-Bianchini F,
Hreglich S and Verita M 1993 NUCL.Instrum. Methods
B 16 215
181. Colombin L. Jelli A. Riea I. Pireaux J J and Verbist 1
.
1977 J. Non-Cryst. &ids' 24 253
[91 Colombin L, Charlier H, lelli A, Debras G and Verbist 1
1980 1. Non-Cyst. Solids 38/39 749
1101
.
. Baitinser W E. French P W and Swarts E L 1980
J. i o n . C y s f . Solids 38/39 749
[l 11 Verita M, Geotti-Bianchini F. Hreglich S, Pantano S G and
Bojan V 1992 BoL Soc. Espana. Cemin. Wd. 31-C 415
[I21 Yang X H and McKeever S W S 1988 Nucl. Tracks
Radiar. Meas. 14 75; 1990 J. Phys. D: Appl. Phys.23
237; 1990 Radiat. Prot. Dosim. 33 27
1131 Song K S and Williams R T 1993 Self Trapped Excilons
(Berlin: Springer)
[141 Khanlary M R, Townsend P D and Townsend J E 1993
J. Phys. 0: Appl. Phys. 26 371
1151 Cazaux J 1986 J. Appl. Phys. 59 1418